Neural-Network Computation Using Spin-Wave-Coupled Spin-Torque Oscillators

A neural-network computation scheme is proposed based on a perceptron model having processing units that consist of spin-wave-coupled spin-torque oscillators. This is an oscillatory neural network, where the relative phase of the oscillators is controlled by tuning the Dzylaoshinshkii-Moriya interaction and applying an oscillating magnetic field. Each processing unit receives an input signal that is an oscillating magnetic field and transmits alternating current. The alternating current is a function of the relative phase and generates an oscillating magnetic (Oersted) field around the wire through which the current flows. The generated Oersted field then becomes an input signal to the next processing unit. By solving the Landau-Lifshitz-Gilbert equation, we obtain an activation function of the processing unit. Finally, an artificial neural network is constructed using the obtained activation function to recognize the handwritten digits in the Modified National Institute of Standards and Technology database.